Quantitative Risk Modeling

Quantitative Risk Modeling

Signal.Engine integrates a quantitative risk layer that sits between the Deep Learning Brain and the Execution Engine. Instead of relying solely on directional predictions, the system evaluates every signal through a statistical lens to ensure capital preservation and mathematical expectancy.

Overview

The risk module (internally referred to as the QuantExpert) calculates the statistical viability of a trade. It transforms a categorical prediction (BUY/SELL/HOLD) into a risk-adjusted decision by calculating Expected Value (EV) and Value at Risk (VaR) based on current market volatility.

Core Risk Metrics

The system exposes a QuantRisk object for every asset scanned. These metrics are used by the SimulationEngine and the live trader_alpaca.py script to determine if a trade meets the minimum safety threshold.

| Metric | Type | Description | | :--- | :--- | :--- | | WinRate | float | The historical probability of success for the current pattern (0.0 - 1.0). | | EV | float | Expected Value: The mean projected return per trade, accounting for both win probability and loss magnitude. | | VaR95 | float | Value at Risk (95%): The maximum expected loss over a specific time horizon at a 95% confidence level. | | MaxDrawdown | float | The peak-to-trough decline observed in the current strategy regime. |

Market Regime Adaptation

The risk model is dynamic and adjusts its sensitivity based on the detected Market Regime. This logic is handled during the brain's "thinking" phase:

• CALM Regime: Standard VaR thresholds. The agent prioritizes trend-following and higher position sizing.
• VOLATILE Regime: Aggressive VaR expansion. The agent triggers "High Volatility" rationales, increases the distance of Stop Losses, and reduces individual trade quantity.

# Internal Logic Example: Regime Detection
vol_count = sum(1 for d in decisions if "High Volatility" in d.get('Rational', []))
if vol_count > len(decisions) * 0.3:
    current_regime = "HIGH_VOLATILITY"

API Usage & Data Structure

The risk metrics are accessible via the /api/results endpoint. Frontend components use this data to color-code signal confidence and block high-risk executions.

Request

GET /api/results

Response Interface (TypeScript)

interface Decision {
    Ticker: string;
    Action: string;
    Confidence: number;
    Rational: string[];
    QuantRisk: {
        WinRate: number;
        EV: number;       // Positive EV is required for 'BUY' signals
        VaR95: number;    // Used to calculate Stop-Loss levels
        MaxDrawdown: number;
    };
}

Sample JSON Output

{
  "Ticker": "RELIANCE.NS",
  "Action": "BUY",
  "Confidence": 0.88,
  "QuantRisk": {
    "WinRate": 0.64,
    "EV": 1.25,
    "VaR95": -2.10,
    "MaxDrawdown": 5.4
  },
  "Rational": ["Sequence Trend Confirm", "Low Volatility Regime"]
}

Implementation Guidelines for Traders

When deploying the agent via src.trader_alpaca, the quantitative layer acts as a gatekeeper:

1. EV Filtering: If EV <= 0, the agent will default to a HOLD or NEUTRAL position, even if the LSTM model predicts an upward move.
2. VaR-Based Sizing: Position sizes are inversely proportional to VaR95. If the potential 95% loss exceeds the account's maximum risk-per-trade (set in .env), the trade quantity is automatically scaled down.
3. Stop-Loss Alignment: The system automatically places Alpaca stop_loss orders at the price point corresponding to the VaR95 level to ensure statistical exits.

Configuration

Risk parameters (such as maximum drawdown limits and base quantity) can be configured via environment variables or CLI flags during deployment:

# Example: Deploy with a specific quantity, subject to QuantRisk scaling
python -m src.trader_alpaca --symbol TCS.NS --qty 10 --live